Visible Digital Image Watermarking by Compound MappingAlgorithm

EVA TUBAUniversity of BelgradeFaculty of Mathematics

Studentski trg 16, BelgradeSERBIA

etuba@ieee.org

MARKO BEKOUniv. Lusofona de Humanidades e Tecnologias

Computer Eng. Dept.Campo Grande 376, Lisbon

PORTUGALmarko@isr.ist.utl.pt

MILAN TUBAJohn Naisbitt University

Graduate School of Comp. Sci.Bulevar umetnosti 29, Belgrade

SERBIAtuba@ieee.org

Abstract: Digital images are used in almost every area of the life. The advent of digital images has broughta lot of progress to various scientific areas, but also many new threads have been opened. One of the mainproblems with digital data and images is unauthorized replication and use. Digital watermarking is one of thetechniques for protecting copyrights of digital images by embedding the additional data to indicate the ownersof the copyright. In this paper a method for visible digital image watermarking based on compound mappingalgorithm is proposed. Compound mapping was used for lossless digital watermarking embedding and removingwhile security was improved by defining parameters of the compound mapping.

Key–Words: Visible image watermarking, compound mapping, security, digital images

1 IntroductionInternet and multimedia technologies have becomepart of everyday life as well as creating and usingdifferent digital data such as texts, images, videos oraudio formats. Massive use of digital data enabledgreat benefits in different areas such as biology [23],medicine [28], [24], astronomy [10], etc. Nowadays,numerous algorithms of image processing and appli-cations based on it can be found. Some of the exam-ples are handwritten digit recognition [30], face detec-tion [11], lip detection [4], leaf recognition [21], im-age segmentation [33], [2], JPEG compression [34],retinal image registration [31], etc.

One of the main issue is unauthorized replica-tion of the digital images. As a result of the problemof unauthorized replication of materials, stenographycontributed to the development of watermarking, away to protect them and preserve their authenticity[14]. Both of them rely on the concept of incorporat-ing information into cover data media. Digital water-mark protects the image from unauthorized copyingand reproduction by embedding the original with ad-ditional data to indicate the owners of the copyright[1].

The three main factors that impact the overallquality of the copyright pattern are robustness, imper-ceptibility and capacity.

• Robustness: The removal or destruction of awatermark should be challenging, for robust isa method of watermark immunity that is de-signed to safeguard images against alteration andmanipulation such as filtering, rotation, sailing,cropping, resizing etc.

• Imperceptibility: Implies that the quality of theoriginal image should not be consumed by theexistence of a watermark.

• Capacity: Incorporates procedures that help fa-cilitate the embodiment of information. [8]

In addition, there are two main kinds of water-marks, visible and invisible. Visible watermarks [12],like those on company logos, are noticeable, that is,they are the first line of defense whose task is toclearly indicate that the material is copyrighted andprevent further violation while invisible watermarkscannot be easily noticed and they are situated in theperiphery of the content and they can be used to vali-date ones copyright.

Image watermarking can be done in spatial or fre-quency domain. In the spatial domain an image is rep-resented in form of a pixelated matrix [26]. Embed-ding of a watermark is done by directly changing thevalue of pixels, which implies a change of the corre-

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 76 Volume 16, 2017

sponding color and other parameters of the pixel ex-actly in the area intended for the insertion of the wa-termark. These algorithms are simple to understandand use as well as less time consuming and can befurther used on any type of image.

On the other hand, in the frequency domain theimage is represented in the form of frequency [19].Initially the original image is converted by a prede-fined transformation and then the watermark is em-bedded into the transformed image or into the trans-formation coefficients and finally, the inverse trans-form is performed to obtain the watermarked im-age. Most commonly used transformations are dis-crete cosine transform (DCT), discrete wavelet trans-form (DWT) and discrete Fourier transform (DFT).

It is important to highlight that the significant partof the well watermarking process is extracting the wa-termark from the watermarked image. The extractionprocess implies the removal of watermark from an al-ready watermarked image. This is a critical part ofthe entire process and that is exactly why security isalways a question of the highest importance. [17]

Thanks to the freedom that internet provides, weare more vulnerable then ever from numerous of dif-ferent kinds of attacks. One of the problems of theseattacks is that by removing a watermark the attacker isable to facilitate distribution, commercialization, andother kinds of malversation. Therefore, it is essentialthat security should entail the protection against dataalteration from various unauthorized users and attack-ers [9].

In order to stop data modification and to help toprotect data from unapproved hackers and end-users,data security is mandatory. As a result of the ex-panding rise in data distribution rate over the inter-net, data security has acquired higher awareness overthe years. Therefore, the development of internet se-curity was necessary which would not only providesecurity against illegal use but safeguard against theattacks that would jeopardize the security as well. Toprevent these kinds of attacks some methods and tech-niques were developed such as cryptography, stenog-raphy and digital image watermarking [27].

In this paper a method for digital image water-marking by securing compound-mapping algorithm ispresented. This eas tested on various images differ-ent in sizes and quality. Quality and security of pro-posed algorithm was measured with several standardmetrics.

The rest of the paper is organized as follows. InSection 2 literature review of algorithms for embed-ding and extracting watermark is presented. In Sec-

tion 3 our proposed algorithm for security improve-ment of algorithms for spatial domain is proposed.Section 4 contains the results obtained by our pro-posed algorithm. And finally Section 5 delivers a con-clusion.

2 Literature ReviewNowadays many different algorithms were proposedfor digital image forgery detection [32]. One of themethods for image protection is embedding digitalwatermark. Today there is a numerous of techniquesfor image watermarking [3]. A great number of thesetechniques is widely used and some of them will bediscussed in this section. However, a number of se-curity attacks and vulnerabilities are on the rise andtherefore there is a lot of space for improvement whenit comes to watermarking. Digital watermarking is al-ready widely used all over the world and many differ-ent approaches have been successfully used, but grow-ing security attacks and the growing need for the useof this system on the increasing number of roads makethis problem still popular for researchers.

In the [13] a simple method for digital imagewatermarking was presented. This method relies onEXIF metadata (camera model, date taken, copyright,etc.) which works as a watermark [16]. Even whenthe marked image has been intentionally modified, theoriginal EXIF with selected information can mostlybe recovered from the channel decoding process. Be-cause it is easy to manipulate and modify EXIF data,this type of algorithm is less secure than others.

A more sophisticated and secure method for dig-ital image watermarking, that relies on frequency do-main was shown in [15]. It was based on joint DWT-DCT [7]. Against frequent signal processing attacks amore advanced robustness, along with imperceptibly,was maintained. Initially, in certain sub-bands of a3-level DWT transformed of a host image, a binarywatermarked image was embedded. Next, the PN-sequences of the watermark bits and the DCT trans-form of each selected DWT sub-band, were insertedin the coefficients of the matching DCT middle fre-quencies. During extraction stages, which maybe at-tacked, the watermarked image was first preprocessedthrough the use of sharpening and various kinds offilters. The similar approach, to the embedding pro-cess, was used to obtain the DCT middle frequen-cies of each sub-band. Ultimately, an interaction wascomputed between PN-sequences and mid-band co-efficients in order to determine watermarked bits. In

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 77 Volume 16, 2017

conclusion, the results showed that the recommendedmethod enhanced the performance of the watermark-ing algorithm which rely on the joint of DWT-DCT.

Two lossless visible watermarking algorithmswere proposed in [35], pixel value matching algorithm(PVMA) and pixel position shift algorithm (PPSA).Since the embedding distortion of noticeable water-marking was generally greater than that invisible wa-termarking, lossless property was highlighted so as topreserve the fidelity of the signal following the wa-termark extraction. In PVMA, the objective intensitymapping function was used in order to watermark avisible logo. PPSA, on the other hand, made use ofcircular pixel shift so as to improve the visibility ofthe watermark in regions with high variance.

In addition to the PVMA and PPSA, there is anolder and easier method of embedding the watermarkknown as least significant bit (LSB) [5]. The im-plementation of a watermark in the LSB pixels doesnot only allow easier implementation but it does notseverely distort the image as well, although whenit comes to attacks it lacks robustness against them.Through the selection of a subset of image pixels andthe substitution of the least important bit of each ofthe selected pixels with the watermark bits, the em-bedding of a watermark is achieved. Furthermore, thewatermark can be expanded all over the image or itmay be located in the preferred positions of the im-age. However, as a result of these primitive techniquesbeing not only vulnerable to attacks but highly sensi-tive to noise and common signal processing, the wa-termark can be easily destroyed.

A security improvement to the mentioned LSBtechnique is the use of a random key which is ob-tained by a pseudo-randomized number generation[25]. Random key is not a final security measure. Itmakes the removal of a watermark more difficult butnot impossible.

In [20] a method which provides visible wa-termarking that does not impact the image quality,named compound mapping method. The proposedmethod used deterministic one-to-one mappings ofimage pixel values where f converts a set of numer-ical values of pixels P = p1, p2, . . . ,pm to another setW = w1, w2, . . ., wm that represents pixels with wa-termark on themselves. The original image can be re-covered from previously watermarked image by usingthe corresponding reverse method. In order to main-tain solid security level, set W was randomized there-fore the extraction of a watermark without a key wasimpossible.

An innovative use of compound capping method

for images in PNG format was presented in [18]. Ad-vantage of the PNG format is that it has the additionalA-alpha channel. In the A-alpha channel plane, thosepixels which underly the ones that have been water-marked in the color channel plane will be appointeda unique alpha value. An additional advantage of thismethod was the fact that during the recovery process,the original watermark image was not required. Inaddition to all this, random generated numbers wereused as a security measure to prevent the removal ofthe watermark without a key.

The most common types of attacks, in the dig-ital image watermarking, are known as removal at-tacks. Extraction attackers endeavor to isolate the wa-termarked data from the data of the original imageand remove only the watermark image, via the anal-ysis and estimation of the watermark. Some examplesof these kinds of attacks may include: collision at-tacks, compression attacks, certain filter operation ordenoising [29].

One way to maintain security is through the useof secret keys, not only for detection but embeddingprocesses as well. The three types of these keys are:private-key, detection-key and public-key. The pur-pose of these keys is to prevent the security attackscovered above, and provide the removal of the water-mark only to the authorized personal that are in pos-session of the keys [6].

In [22] literature review of reversible digital wa-termarking techniques is given.

As mentioned above, there are very successfulmethods of image watermarking, but there is a lot ofspace for improvement. In the next section we pro-pose our method for better security protection in theprocess of embedding and extracting watermark.

3 Our Proposed MethodOne to one compound mapping represents rather newmethod for visible watermarking with a capability oflossless image recovery. The method is based on theuse of one to one compound mapping of image pix-els value and obtains value close to pixels of the de-sired visible watermarks. Compound mapping meth-ods usually use randomized keys in the process of wa-termark embedding. There is no precise explanationfor the exact algorithm used to generate random num-bers. In this paper, we propose using some concretealgorithm for generating pseudo-randomized numberswhich will add more complexity to the numbers andkeys which will disable the removal of the watermark

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 78 Volume 16, 2017

due to the mentioned complexity.An algorithm based on compound mapping for

embedding different types of watermarks will be pre-sented. The original image can be recovered fromprotected image without any loss by using reversemapping. This approach can be used to embedmonochrome and color watermark into color images.

In general, one to one compound mapping canbe defined as follows. Assume a set of values A ={a1, a2, a3, . . . , am} is converted into another setB = {b1, b2, b3, . . . , bn} and that the mapping fromai to bi is reversible where ai and bi stand for the im-age pixel values. The compound mapping f is definedby a one-to-one function Fx with one parameter x = por q:

b = f(a) = F−1q (Fp(a)) (1)

where F−1x is the inverse Fx. Based on one-to-oneproperty, if Fp(a

′) = a then F−1p (a) = a′ for all val-ues of a and p. On the other hand, Fp(a) and Fq(a)are set to be unequal if p 6= q.

In our proposed algorithm one to one compoundmapping is used for embedding and extracting the wa-termark to an image. If color watermark is used, themappings can be applied to each of the color channelsand obtain multiple independent results. The result-ing visible watermark is represented by the compositeresult.

Usage of the compound mapping for embeddingor extracting the watermark to an image will be ex-plained by using pseudo code.

Algorithm 1: Process of watermark embed-dingInput: Color original (host) image I, a watermark L.Output: Watermarked Image W.

1. Select a set P of pixels from I where W is to beembedded. The set of pixel P depicts the val-ues of pixels prior the watermark embedding pro-cess.

2. Designate the set of pixels equivalent to P in Wby Q, where Q is being the set of pixels whichincludes the values of the watermark pixels.

3. For each pixel X within p in P, designate theequivalent pixel in Q as Z, the value of the equiv-alent pixel y in L as l, and conduct the followingsteps:

(a) Obtain value a to be a value close to p, viathe utilization of the values of the neighbor-ing pixels of X (excluding X itself).

(b) Appoint b to be the value l

(c) Map p to a new value

q = F−1b (Fa(p))

(d) Set the value of Z to be q.

4. Set the value of each remaining pixel in W, whichis outside the region P, to be equal to that of thecorresponding pixel in l.

Our proposed method includes pseudo-randomized numbers into Algorithm 1 in orderto improve a security. The goal is to get morecomplex key on the part when pseudo-randomizednumbers are generated so that can not be easilybroken, which would lead to watermark removing.

Algorithm 2: Process of watermark extraction

Input: Watermarked image W, and a Water-mark LOutput: Original (hosted) image R, recovered fromW

1. Select the exact watermarking are Q in W as thatwhich was selected in Algorithm 1 (Indicates thearea where the watermark is located).

2. Appoint the value of each pixel inR, which is outside the region Q, to be identicalto that of the corresponding pixel in W. Pixelsoutside the watermarked area are not changed,and thus retain the same values.

3. For each pixel Z with the value q in Q, designatethe equivalent pixel in the recovered image R asx and the value of corresponding pixel Y in L asl, and conduct the following steps:

(a) Attain the same value a as that derived instep 3a of Algorithm 1 by administering thesame estimation technique that was usedthere.

(b) Set b to be the value l.

(c) Restore p from q by setting

p = F−1a (Fb(q))

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 79 Volume 16, 2017

(d) Set the value of x to be p.

In our proposed algorithm for value b was usedwatermark pixel value. For value of parameter awhich should be close to the value of the pixel value pthat should be mapped was used average value of theneighbor pixels. Function Fa(p) was define as:

Fa(p) = (p− a)mod 256 (2)

while

F−1a (p) = (p+ a)mod 256. (3)

4 Experimental resultsThe proposed algorithm was implemented using Mat-lab R2016a and all experiments were performedon the platform Intel R© CoreTMi7-3770K CPU at4GHz, 8GB RAM, Windows 10 Professional OS.Our proposed algorithm for digital image water-marking based on compound mapping algorithmwas tested on a standard test image widely usedin a field of image processing. Results are pre-sented for two different test images: Lena, Man-drill and Cameraman. Both images can be found athttps://homepages.cae.wisc.edu/∼ece533/images/.

The first example is when gray scale watermark isembedded into gray scale image (Cameraman). Orig-inal image is shown in Fig. 1. In Fig. 2 digital water-mark was embedded into the image. Watermark wasremoved and the result image showed in 3 is identicalto the original image. This is the simplest examplessince no combination of color component is neces-sary. Following examples has color host image, thusthe mappings was applied to each of the color chan-nels (RGB color model) and multiple independent re-sults were obtained. The resulting visible watermarkis represented by the composite result.

Compound mapping method with our proposedmethod for security improvement proved to be effi-cient for image watermarking; simultaneously for em-bedding and extracting watermarks. In Fig. 4 is thepreviously mentioned standard color test image. Thisimage is used as a host image for embedding water-mark.

In Fig. 5, the same image is presented with anembedded watermark using the formerly mentionedcompound mapping algorithm with our proposed im-proved security. The process of watermark embeddingwas explained in Algorithm 1, which is further definedvia the pseudo-code.

Figure 1: Original image

Figure 2: Image with watermark embedded

Figure 3: Image with watermark removed

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 80 Volume 16, 2017

Figure 4: Original image

Figure 5: Image with watermark embedded

Figure 6: Image with watermark removed

Figure 7: Original image

Figure 8: Image with watermark embedded

Figure 9: Image with watermark removed

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 81 Volume 16, 2017

Figure 6 represents a watermarked image afterthe watermark extraction process, which can be like-wise explained via the pseudo-code. As we can see,the quality of the image, which had gone through theprocesses of embedding and extraction, is almost thesame as the original image showed in Fig. 1. Hav-ing tested our proposed algorithm on fifteen images,a conclusion was reached that the algorithm not onlyhad good performances but it also preserved the orig-inal quality.

Similar to the previous example, the same water-mark was embedded to the Mandrill image. As it canbe seen image quality remains the same. Resultingimages for this example are shown in Fig. 7, 8 and 9.

5 ConclusionSecurity improvement of the compound mapping al-gorithm used both for embedding and extraction, wasproposed in this paper. Security improvement is re-lated to the parts of embedding as means of generat-ing a secret key to prevent watermark removal via thevarious attacks. Thus, the watermark extraction pro-cess can only be successful with the use of the correctsecret key. Compound mapping algorithm with ourimprovements was tested on standard benchmark im-ages and the predicted results were met. Having inmind that we have discussed about the algorithm thatworks with the spatial domain, future research mayinclude improvement of security in frequency domainand invisible watermarking method.

Acknowledgments: This research is supported byMinistry of Education, Science and Technological De-velopment of Republic of Serbia, Grant No. III-44006.

References:

[1] M. Abdullatif, A. M. Zeki, J. Chebil, and T. S.Gunawan, “Properties of digital image water-marking,” in Signal Processing and its Applica-tions (CSPA), 2013 IEEE 9th International Col-loquium on. IEEE, 2013, pp. 235–240.

[2] A. Alihodzic and M. Tuba, “Bat algorithm (BA)for image thresholding,” Recent Researches inTelecommunications, Informatics, Electronicsand Signal Processing, pp. 364–369, 2013.

[3] T. K. Araghi, A. B. A. Manaf, M. Zamani, andS. K. Araghi, “A survey on digital image wa-termarking techniques in spatial and transformdomains,” International Journal of Advances inImage Processing Techniques, vol. 3, no. 1, pp.6–10, 2016.

[4] A. Arsic, M. Jordanski, and M. Tuba, “Improvedlip detection algorithm based on region segmen-tation and edge detection,” in 23rd Telecommu-nications Forum (TELFOR). IEEE, 2015, pp.472–475.

[5] A. Bamatraf, R. Ibrahim, M. Salleh, andN. Mohd, “A new digital watermarking al-gorithm using combination of least signifi-cant bit (LSB) and inverse bit,” arXiv preprintarXiv:1111.6727, 2011.

[6] M. M. U. Bhaisare and V. R. Raut, “Genericlossless visible watermarking,” InternationalJournal of scientific research and management(IJSRM), vol. 3, no. 4, pp. 2571–2581, 2015.

[7] K. Deb, M. S. Al-Seraj, M. M. S. Kowsar,and I. H. Sarkar, “A joint DWT-DCT basedwatermarking technique for avoiding unautho-rized replication,” in 7th International Forum onStrategic Technology (IFOST). IEEE, 2012, pp.1–5.

[8] B. L. Gunjal and R. Manthalkar, “An overviewof transform domain robust digital image wa-termarking algorithms,” Journal of EmergingTrends in Computing and Information Sciences,vol. 2, no. 1, pp. 37–42, 2010.

[9] P. Gupta, “Cryptography based digital imagewatermarking algorithm to increase security ofwatermark data,” International Journal of Scien-tific & Engineering Research, vol. 3, no. 9, pp.1–4, 2012.

[10] Y.-Q. Han, Y.-Q. Bai, and F.-k. Jia, “Extractingthe centroid from the sun-moon image in astron-omy orientation,” in 33rd Chinese Control Con-ference (CCC). IEEE, 2014, pp. 4721–4726.

[11] R. C. Hrosik, M. Tuba, and M. Vukovic, “Facedetection algorithm based on skin detectionand invariant moments,” in Recent Advances inKnowledge Engineering and Systems Science,2013, vol. 10, pp. 110–115.

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 82 Volume 16, 2017

[12] F.-H. Hsu, M.-H. Wu, C.-H. Yang, and S.-J.Wang, “Visible watermarking with reversibil-ity of multimedia images for ownership declara-tions,” The Journal of Supercomputing, vol. 70,no. 1, pp. 247–268, 2014.

[13] H.-C. Huang and W.-C. Fang, “Metadata-basedimage watermarking for copyright protection,”Simulation Modelling Practice and Theory,vol. 18, no. 4, pp. 436–445, 2010.

[14] M. Hussain and M. Hussain, “A survey ofimage steganography techniques,” InternationalJournal of Advanced Science and Technology,vol. 54, pp. 113–124, 2013.

[15] S. A. Kasmani and A. Naghsh-Nilchi, “A newrobust digital image watermarking techniquebased on joint dwt-dct transformation,” in hirdInternational Conference on Convergence andHybrid Information Technology (ICCIT), vol. 2.IEEE, 2008, pp. 539–544.

[16] E. Kee, M. K. Johnson, and H. Farid, “Digitalimage authentication from jpeg headers,” IEEETransactions on Information Forensics and Se-curity, vol. 6, no. 3, pp. 1066–1075, 2011.

[17] A. Khan, A. Siddiqa, S. Munib, and S. A. Ma-lik, “A recent survey of reversible watermarkingtechniques,” Information sciences, vol. 279, pp.251–272, 2014.

[18] C.-W. Lee and W.-H. Tsai, “A new lossless visi-ble watermarking method via the use of the PNGimage,” 2012.

[19] G. D. Leena, S. S. Dhayanithy, and M. Hwang,“Robust image watermarking in frequency do-main,” International Journal of Innovation andApplied Studies, pp. 2028–9324, 2013.

[20] T.-Y. Liu and W.-H. Tsai, “Generic lossless vis-ible watermarkinga new approach,” IEEE trans-actions on image processing, vol. 19, no. 5, pp.1224–1235, 2010.

[21] M. Lukic, E. Tuba, and M. Tuba, “Leaf recog-nition algorithm using support vector machinewith Hu moments and local binary patterns,”in 15th International Symposium on AppliedMachine Intelligence and Informatics (SAMI).IEEE, 2017, pp. 000 485–000 490.

[22] N. Narawade and R. Kanphade, “Reversible wa-termarking: a complete review,” InternationalJournal of Computer Science and Telecommuni-cations, vol. 2, no. 3, pp. 46–50, 2011.

[23] K. S. P. S. Nellore and P. M. SWAMY, “Genericlossless visible watermarking,” InternationalJournal of Engineering Science & Management,vol. 3, no. 2, pp. 1–5, 2013.

[24] M. Nikolic, E. Tuba, and M. Tuba, “Edge de-tection in medical ultrasound images using ad-justed Canny edge detection algorithm,” in 24thTelecommunications Forum (TELFOR). IEEE,2016, pp. 691–694.

[25] P. K. Sharma, “Rajni: Analysis of image wa-termarking using least significant bit algorithm,”International Journal of Information Sciencesand Techniques (IJIST) Vol, vol. 2, 2012.

[26] A. K. Singh, N. Sharma, M. Dave, and A. Mo-han, “A novel technique for digital image wa-termarking in spatial domain,” in 2nd IEEE In-ternational Conference on Parallel Distributedand Grid Computing (PDGC). IEEE, 2012, pp.497–501.

[27] P. Singh and R. Chadha, “A survey of digitalwatermarking techniques, applications and at-tacks,” International Journal of Engineering andInnovative Technology (IJEIT), vol. 2, no. 9, pp.165–175, 2013.

[28] A. Stojak, E. Tuba, and M. Tuba, “Frameworkfor abnormality detection in magnetic resonancebrain images,” in 24th Telecommunications Fo-rum (TELFOR). IEEE, 2016, pp. 687–690.

[29] H. K. Sunesh, “Watermark attacks and applica-tions in watermarking,” in National Workshop-Cum-Conference on Recent Trends in Mathe-matics and Computing (RTMC), 2011.

[30] E. Tuba and N. Bacanin, “An algorithm for hand-written digit recognition using projection his-tograms and SVM classifier,” in 23rd Telecom-munications Forum Telfor (TELFOR). IEEE,2015, pp. 464–467.

[31] E. Tuba, M. Tuba, and E. Dolicanin, “Adjustedfireworks algorithm applied to retinal image reg-istration,” Studies in Informatics and Control,vol. 26, no. 1, pp. 33–42, 2017.

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 83 Volume 16, 2017

[32] I. Tuba, E. Tuba, and M. Beko, “Digital im-age forgery detection based on shadow texturefeatures,” in 24th Telecommunications Forum(TELFOR). IEEE, 2016, pp. 679–682.

[33] M. Tuba, “Multilevel image thresholding bynature-inspired algorithms-a short review,” TheComputer Science Journal of Moldova, vol. 22,no. 3, pp. 318–338, 2014.

[34] M. Tuba and N. Bacanin, “JPEG quantization ta-

bles selection by the firefly algorithm,” in Inter-national Conference on Multimedia Computingand Systems (ICMCS). IEEE, 2014, pp. 153–158.

[35] S.-K. Yip, O. C. Au, C.-W. Ho, and H.-M.Wong, “Lossless visible watermarking,” in 2006IEEE International Conference on Multimediaand Expo. IEEE, 2006, pp. 853–856.

WSEAS TRANSACTIONS on SYSTEMS Eva Tuba, Marko Beko, Milan Tuba

E-ISSN: 2224-2678 84 Volume 16, 2017